STRIKE heat flux reconstruction by neural networks: application to the experimental results

To characterize the SPIDER negative ion beam in terms of beam uniformity and divergence during short pulse operations an instrumented calorimeter named STRIKE (Short-Time Retractable Instrumented Kalorimeter Experiment) has been designed and operated. STRIKE is made of 16 1D Carbon Fibre-Carbon Composite (CFC) tiles, intercepting the whole beam and observed on the rear side by two infrared (IR) cameras. To know the energy flux profile hitting the front surface and then the beam parameters, it is necessary to solve an inverse non-linear problem, mathematically ill-posed, upon knowing the non-linear characteristics of the tiles and the 2D temperature pattern measured on the rear side of the tiles themselves. Most of the conventional methods used to solve this inverse problem are unbearably time consuming; when fully operative STRIKE receives 1280 beamlets each one characterized by at least 5 features, so a ready-to-go tool to determine the beam condition, is highly recommended. In this work, the inverse problem, in stationary conditions, is faced by using a Neural Network (NN) model, pursuing different training approaches. The NN is trained by associating features extracted from the 2D temperature profile, retrieved from the frequency domain of the camera frames or obtained by a fitting process, to the heat flux profile parameters. The proposed method is then applied to experimental STRIKE data from the beam campaigns, showing very good agreement with calorimetric measurements; particularly the dependence of the beam features on the operational parameters is quantified.